refine current sphinx docs

[pfliu-nlp]

• We can learn from comet: https://github.com/Unbabel/COMET/tree/master/docs/source
  It seems comet's is more powerful, also using more extensions.

• We can detail some necessary information for each metric: https://github.com/ExpressAI/EaaS_API_dev/blob/main/docs/source/metrics.rst

[yyy-Apple]

高亮文字